Hannah Christensen (she/her)

Abstract:

It is often difficult to attribute uncertainty and errors in atmospheric models to designated model components. This is because sub-grid parameterised processes interact strongly with the large-scale transport represented by the explicit model dynamics. We carry out experiments with prescribed large-scale dynamics and different sub-grid physics suites. This dataset has been constructed for the Model Uncertainty Model Intercomparison Project (MUMIP), in which each suite forecasts sub-grid tendencies at a 22km grid. The common dynamics is derived from a convection-permitting benchmark: an ICON DYAMOND experiment (2.5km grid). We compare four different physics suites for atmospheric models in an Indian Ocean experiment. We analyse their joint PDFs of precipitation and associated physics tendencies for a full month, where precipitation is used to diagnose uncertainty of convective activity. We find that all physics suites produce very similar precipitation amounts, with very high correlations between models, i.e.,  > 0.95 at the native grid. However, the convection-permitting benchmark is more dissimilar from each of the physics suites, with correlations of  ≈ 0.80. Similarly, we show that the vertically averaged physics tendencies in the free-troposphere are highly similar between the four physics suites, yet different if reconstructed for the benchmark. The water vapour sink is very closely linked with precipitation in the four physics suites. This suggests that the coarse-grid models are overconfident. We hypothese is that variation in unresolved convective structures can lead to variation in the dynamics, following a given amount of latent heating at fine grids, but not in our physics suites. The difference appears to be caused by the explicit interactions between gravity waves, occurring at fine grids only Groot et al. (2024) We assess whether their non-linear feedback from convective precipitation systems explains our joint PDFs of precipitation. The slightly exponential curve supports the interaction mechanism. These findings are further evidence for a non-linear feedback between convective organisation/aggregation and dynamics. This feedback has been studied earlier in a real-case study with ICON by looking from the fixed-physics rather than the fixed-dynamics perspective. Our current results may indicate that sub-grid physics with stochastic physics perturbations emulate convective organisation effects.

Abstract:

Abstract. The Atlantic Multidecadal Variability (AMV) and the North Atlantic Oscillation (NAO) are the dominant modes of oceanic and atmospheric variability in the North Atlantic, respectively, and are key sources of predictability from seasonal to decadal timescales. However, the physical processes and feedback mechanisms linking the AMV and NAO, and the role of diabatic processes in these feedbacks, remain debated. We present a data-driven dynamical modelling framework which captures coupled decadal variability in AMV, NAO, and North Atlantic precipitation. Applying equation discovery methods to observational data, we identify low-order models consisting of three coupled ordinary differential equations. These models reproduce observed decadal variability and show robust out-of-sample predictive skill on multi-annual to decadal lead times. The resulting model dynamics include a distinct quasi-periodic 20-year oscillation consistent with a damped oceanic mode of variability. Notably, precipitation-related terms feature prominently in the low-order models, suggesting an important role for latent heat release and freshwater fluxes in mediating ocean–atmosphere interactions. We propose new feedback mechanisms between North Atlantic sea surface temperature and the NAO, with precipitation acting as a dynamical bridge. Overall, these results illustrate how equation discovery can provide mechanistic hypotheses and new insight beyond conventional analyses of observations and climate model simulations.

Abstract:

Abstract Next-generation high-resolution (km-scale) climate models promise unprecedented accuracy in climate projections, but realizing their potential requires robust methods to quantify how well simulations align with real-world observations. Average-based metrics conventionally used for climate model evaluation ignore the physics encoded in the fine-scale structures of km-scale simulations. To overcome this limitation, we propose a novel, statistically principled evaluation methodology based on the likelihood function of a generative image model. Our method provides a continuous similarity metric derived from the likelihood distribution of observation and simulation snapshots, which can redefine the evaluation, intercomparison, and parameter tuning of high-resolution climate models. We demonstrate the applicability and interpretability of this method by evaluating convective clouds simulated by two state-of-the-art global km-scale models, using their outgoing infrared radiation fields. This work establishes a scalable pathway toward observation-based evaluation of next-generation climate simulations.

Abstract:

Representing and quantifying uncertainty in physical parameterisations is a central challenge in weather and climate modelling, and approaches are often developed separately for different time‐scales. Here, we introduce a unified framework for analysing uncertainty in parameterisations across weather and climate regimes. Using the Lorenz 1996 system as a testbed for simplified chaotic dynamics, we quantify uncertainties in a subgrid‐scale parameterisation using a Bayesian neural network (BNN). This allows us to disentangle aleatoric uncertainty, arising from internal variability in the training data, and epistemic uncertainties, arising from poorly constrained parameters during training. At runtime, we sample uncertainties in line with stochastic approaches in weather models and perturbed‐parameter methods in climate models. On weather time‐scales, aleatoric uncertainty dominates, underscoring the value of stochastic parameterisations. On longer, climate time‐scales and under changing forcings, accounting for both types of uncertainty is necessary for well‐calibrated ensembles, with epistemic uncertainty widening the range of explored climate states, and aleatoric uncertainty promoting transitions between them. Constraining parameter uncertainty with short simulations reduces epistemic uncertainty and improves long‐term model behaviour under perturbed forcings. This framework links concepts from machine learning with traditional uncertainty quantification in earth system modelling, offering a pathway towards seamless treatment of uncertainty in weather and climate prediction.

Abstract:

Abstract Subgrid machine‐learning (machine learning [ML]) parameterizations have the potential to introduce a new generation of climate models that incorporate the effects of higher‐resolution physics without incurring the prohibitive computational cost associated with more explicit physics‐based simulations. However, important issues, ranging from online instability to inconsistent online performance, have limited their operational use for long‐term climate projections. To more rapidly drive progress in solving these issues, domain scientists and ML researchers opened up the offline aspect of this problem to the broader ML and data science community with the release of ClimSim, a NeurIPS Data sets and Benchmarks publication, and an associated Kaggle competition. This paper reports on the downstream results of the Kaggle competition by coupling emulators inspired by the winning teams' architectures to an interactive climate model (including full cloud microphysics, a regime historically prone to online instability) and systematically evaluating their online performance. Our results demonstrate that online stability in the low‐resolution real‐geography setting is reproducible across multiple diverse architectures, which we consider a key milestone. All tested architectures exhibit strikingly similar offline and online biases, though their responses to architecture‐agnostic design choices (e.g., expanding the list of input variables) can differ significantly. Multiple Kaggle‐inspired architectures achieve state‐of‐the‐art results on certain metrics such as zonal mean bias patterns and global Root Mean Squared Error, indicating that crowdsourcing the essence of the offline problem is one path to improving online performance in hybrid physics‐AI climate simulation. Plain Language Summary Future climate models may use machine learning (ML) to replace small‐scale physical processes that are otherwise too costly to simulate directly over long timescales. Such “hybrid” physics–ML models could improve predictions by reducing uncertainties from current approximations. But making them run reliably in full climate simulations has been a major challenge. To speed progress, scientists created an open data set, benchmarking framework, and global competition to drive improvement for these ML components. This paper follows up on that competition by testing ideas from the winning teams within hybrid climate models. For the first time, we show that stable hybrid simulation is now reproducible across a range of diverse ML architectures. We find that different architectures share similar patterns of errors both before and after coupling, although their responses to added training inputs can differ. Finally, some competition‐inspired designs achieve state‐of‐the‐art scores on individual performance measures, but no single approach beats the previous benchmark (Hu et al., 2025, https://doi.org/10.1029/2024ms004618 ) on every metric. Key Points Online stability in the low‐resolution real‐geography setting is reproducibly achievable across diverse architectures Offline and online zonal mean biases are near‐identical across architectures; online runs underestimate tropical precipitable water An expanded variable list is universally beneficial offline but has diverging, architecture‐dependent effects online